Generalizing Group Fairness in Machine Learning via Utilities